check_heartbeat (d7cc82c) - check_heartbeat

check_heartbeat

#!/bin/bash
# Author: Benjamin Renard
# Date: 2011-10-26
# Description: Nagios check plugin to test Linux HA cluster
#              status using cl_status

# Get program path
REVISION=1.0
PROGNAME=`/usr/bin/basename $0`
PROGPATH=`echo $0 | /bin/sed -e 's,[\\/][^\\/][^\\/]*$,,'`

NODE_NAME=`uname -n`
CL_ST='/usr/bin/cl_status'

WARNING_NUMBER_PING_NODE=1

if  [ -f $PROGPATH/utils.sh ]
then
	.$PROGPATH/utils.sh
elif [ -f /usr/lib/nagios/plugins/utils.sh ]
then
	. /usr/lib/nagios/plugins/utils.sh
else
	echo "Can't find utils.sh"
	exit 3
fi

function usage {
    echo "Nagios plugin to test heartbeat status using cl_status

Usage: $0 [-h|--help] [-v|--version]
  --help -h	Print this help information
  --version -v  Print version of plugin
"
}

case "$1" in
	--help | -h)
		print_revision $PROGNAME $REVISION
		echo
		usage
		echo
		support
		exit $STATE_OK
	;;
	--version | -v)
		print_revision $PROGNAME $REVISION
		exit $STATE_OK
	;;
esac

$CL_ST hbstatus > /dev/null
res=$?
if [ $res -ne 0 ]
then
  echo "CRITICAL: Heartbeat is not running on this node"
  exit $CRITICAL
fi

NODES=`$CL_ST listnodes`

PERF_DATA=""

NODE_COUNT=0
NODE_ACTIVE=0
PING_COUNT=0
PING_ACTIVE=0
LINK_DOWN=0
for node in $NODES
do
	if [ "$node" == "$NODE_NAME" ]
	then
		let NODE_COUNT=NODE_COUNT+1
		let NODE_ACTIVE=NODE_ACTIVE+1
		continue
	fi
	type=$( $CL_ST nodetype $node )
	status=`$CL_ST nodestatus $node`
	res=$?
	case "$type" in
		normal)
			link=0
			link_up=0
			let NODE_COUNT=NODE_COUNT+1
			if [ "$status" == "active" ]
			then
				let NODE_ACTIVE=NODE_ACTIVE+1
			fi
			for link in $( $CL_ST listhblinks $node )
			do
				let link_count=link_count+1
				status=$( $CL_ST hblinkstatus $node $link )
				if [ "$status" == "up" ]
				then
					let link_up=link_up+1
				else
					let LINK_DOWN=LINK_DOWN+1
				fi
			done
			let warn=link_count-1
			PERF_DATA="$PERF_DATA linkup_$node:$link_up;$warn;0;0;$link_count"
		;;
		ping)
			let PING_COUNT=PING_COUNT+1
			if [ "$status" == "ping" ]
			then
				let PING_ACTIVE=PING_ACTIVE+1
			fi	
		;;
	esac
done

PERF_DATA="$PERF_DATA ping_active:$PING_ACTIVE;$WARNING_NUMBER_PING_NODE;0;0;$PING_COUNT"

let warn=NODE_COUNT-1
PERF_DATA="$PERF_DATA node_active:$NODE_ACTIVE;$warn;0;0;$NODE_COUNT"

msg="$NODE_ACTIVE/$NODE_COUNT heartbeat node(s) active(s)"
if [ $NODE_ACTIVE -eq 0 ]
then
	echo "CRITICAL : $msg |$PERF_DATA"
	exit $STATE_CRITICAL
elif [ $NODE_ACTIVE -ne $NODE_COUNT ]
then
	echo "WARNING : $msg |$PERF_DATA"
	exit $STATE_WARNING
elif [ $LINK_DOWN -ne 0 ]
then
	echo "WARNING : $msg but all link are not up |$PERF_DATA"
	exit $STATE_WARNING
elif [ $PING_ACTIVE -le $WARNING_NUMBER_PING_NODE ]
then
	echo "WARNING : $msg but only $PING_ACTIVE ping node(s) up |$PERF_DATA"
	exit $STATE_WARNING
else
	echo "OK : $msg |$PERF_DATA"
	exit $STATE_OK
fi

check_heartbeat

Initial commit